Dockerfile

# GapGuide backend — Hugging Face Docker Space image.
# Django + DRF API with the in-process NER chain, served by gunicorn on :7860.
FROM python:3.13-slim

ENV PYTHONUNBUFFERED=1 \
    PYTHONDONTWRITEBYTECODE=1 \
    PIP_NO_CACHE_DIR=1 \
    HF_HOME=/app/.cache/huggingface \
    NLTK_DATA=/app/.cache/nltk

# HF Spaces run the container as UID 1000.
RUN useradd -m -u 1000 appuser
WORKDIR /app

# Python deps first for better layer caching. requirements.txt carries the
# torch CPU --extra-index-url, so CPU wheels resolve here (verified: torch
# 2.11/2.12 +cpu cp313 manylinux exist).
COPY requirements.txt .
RUN pip install --no-cache-dir -r requirements.txt \
    && python -m spacy download en_core_web_sm

# App code, owned by the runtime user in a single COPY (no recursive-chown bloat).
COPY --chown=1000:1000 . .

# Make the dirs the next steps write to owned by UID 1000 (the /app dir itself
# is root-owned from WORKDIR; chown only the specific dirs, not the whole tree).
RUN mkdir -p /app/.cache/huggingface /app/.cache/nltk /app/staticfiles \
    && chown 1000:1000 /app /app/.cache /app/.cache/huggingface /app/.cache/nltk /app/staticfiles

USER 1000

# Bake everything the NER chain downloads → fast, offline-safe first parse in prod.
# Runs as UID 1000 so caches land under the owned /app/.cache.
RUN python -m nltk.downloader -d /app/.cache/nltk stopwords punkt punkt_tab \
    && python scripts/prefetch_models.py

# Collect Django admin / DRF static for WhiteNoise. Inline build env sidesteps the
# DEBUG=False "insecure SECRET_KEY" guard; it does not persist to runtime.
RUN DEBUG=True SECRET_KEY=build-only python manage.py collectstatic --noinput

EXPOSE 7860
# /bin/sh form survives a Windows upload (no exec-bit / shebang dependency).
ENTRYPOINT ["/bin/sh", "entrypoint.sh"]